More on pseudospectra for polynomial eigenvalue problems and applications in control theory (Q1611913)

Next to the usual linear spectral (or eigenvalue) problem \(A x = \lambda x\) one finds the generalized one, \(A x = \lambda B x\). Moving on, a square-matrix polynomial problem follows, \(P(\lambda) x = 0\). Its non-square form appeals finally to the authors motivated (say, by the theory of games) to study the ``infinite'' eigenvalues \(\lambda=\alpha/\beta \) with \(\beta = 0\). They contemplate the homogeneous polynomial and add perturbations, \([P(\alpha ,\beta)+ \Delta P'(\alpha,\beta)] x = 0\). In this way the authors define the (complex) pseudospectrum (= a set of pairs (\(\alpha,\beta)\)) and show how it can be calculated via minimization of \(P x\) over \(x\) with unit norm (cf. their main Theorem 2.1). In the sequel they illustrate the concept (on the Wilkinson's pathological example), suggest a vizualization on the Riemann sphere (a projection which treats also the information about the infinite eigenvalues properly) and outline a few applications [studying the distance from the nearest nonregular matrix polynomial or from a domain of uncontrollability (= impossibility to reach some final states) in a dynamical system]. Finally they extend the concept, in the spirit of their preceding paper [SIAM J. Numer. Anal. Appl. 23, No. 1, 187-208 (2001; Zbl 0996.65042)], to the so called structured pseudospectra and illustrate their merits on a damped mass-spring example of \textit{F. Tisseur}'s older paper [Linear Algebra Appl. 309, NO. 1-3, 339-361 (2000; Zbl 0955.65027)].